Testing AMD's Fidelity FX Super Resolution 2.0

How does Fidelity FX Super Resolution 2.0 work? 

Like FSR 1.0, FSR 2.0 lowers the rendering resolution of a game, increasing performance. Then it scales the image back up again to your target resolution, to maintain a near identical image, similar to what you would get without FSR 2.0 enabled.

It uses a technique known as temporal scaling, that uses frame color, depth, and motion vectors in a game rendering pipeline combined with information of past frames to create a high-quality upscaled image at the target resolution.

FSR 2.0 is integrated earlier into a game's rendering pipeline and includes an optimized anti-aliasing solution. By comparison, FSR 1.0 used a spatial upscaling algorithm, that is limited to the game's current frame to draw information from to create an upscaled image. It also relied on a game's implementation of anti-aliasing.

Okay. So how does this effect my performance? 

Given the industry trend to move to more demanding graphical options like Raytracing, gaming at higher resolutions requires a lot of power to run smoothly.

Image upscaling implementations lower the render resolution to almost half, easing the load on the CPU and GPU.

Then using the lower quality image as a template, the image is upscaled back to the original resolution, giving you the performance benefits, in exchange of near identical image quality that you would normally get at that resolution.

This makes it much more viable to run demanding games not only on PC's but on consoles as well.

Cool. So that means I can run any game at max settings? 

Not exactly. Image upscaling solutions like FSR 2.0 might not need dedicated hardware, but you will still be limited by the horsepower of your rig. Naturally, high-end graphic cards and CPUs will gain the most but luckily, there are certain presets you can use on mid-range to lower-end systems to increase performance.

In the case of FSR 2.0, there are three presets you need to be aware of.

Quality mode will scrunch your image down by 67% of your original resolution, which means if your target resolution is 1080p, the image will first be rendered at 720p, before being upscaled back to native resolution or higher.

Naturally, this mode is the lowest in terms of hit on image quality, while giving you decent performance gains.

Balanced mode, renders a 1080p image at 59% of screen resolution, 1129 x 635, before upscaling it back. As the name implies, this is the ideal compromise between image quality and performance.

Performance mode lowers the render resolution down to 50% of the screen resolution, which means a 1080p image is rendered at 540p, before being upscaled back to near native or similar image quality.

AMD recommends higher end Radeon RX 5700 graphic cards and above, or Nvidia's RTX 20-series GPUs and above for FSR 2.0's Quality mode. This is meant for higher-end PC's that target 4K resolution.

For PC's that are built for 1440p resolutions, AMD recommends Radeon RX Vega Series GPUs, Radeon RX 5600 and above, all the way up to RX 6600, or Nvidia's GTX 1080, RTX 2060 and RTX 3060 for FSR 2.0's Balanced mode.

PC's targeting 1080p can use Performance mode with AMD's Radeon RX 590, and Radeon RX 6500XT GPUs or Nvidia's GeForce GTX 1070, or the GeForce GTX 16 Series GPUs.

Benchmarks

Arkane's Deathloop, which we reviewed last year, will be the first game to make use of AMDs FSR 2.0, and we got to test it to see what difference FSR 2.0 makes in those all important framerates.

For the test, we replayed the opening of the game with all settings at the Ultra preset in-game, Vsync turned off and the frame limiter set as high as it would go to 240 fps. The sharpening setting for FSR was set to 10, which is the maximum.

We tested with Raytraced Ambient Occlusion and Sun Shadows enabled for one pass, and disabled for the other. We also ran through all modes in FSR 2.0 - Quality, Balanced and Performance.

We used the Quality preset in FSR 2.0, when testing at 4K, the Balanced preset when testing at 1440p and Performance preset, when testing at 1080p.

Benchmark system

CPU: AMD Ryzen 9 5950X processor @ 3.40GHz

GPU: AMD Radeon XT 6900 with 16GB of VRAM

Memory: Corsair Vengeance 16GB DDR 4 RAM @ 2400MHz

Motherboard: Asus ProArt X570 Creator WiFi 

Hard Drive: Kingston 512GB Internal SSD

At 4K resolution and using the Quality preset in FSR 2.0 settings, we were able to get an average of 75 fps with Raytraced Sun shadows and Raytraced Quality Ambient Occlusion. Turning off Raytracing, didn't seem to impact performance that match, giving us an average of 77 fps.

Using Balanced Mode at a resolution of 1440p, we were able to extract an average fps of 111 with Raytracing options turned on. Turning them off, increased the average to 115 fps.

1080p resolution was tested with FSR 2.0's Performance mode, and with Raytracing turned on, we got an average of 124 fps. With Raytracing off, we recorded an average of 127 fps.

As for image quality, the Quality preset looked nearly indistinguishable to our eye, compared to a normally rendered 4K frame. It's impressive how close the upscaled image gets to the target resolution, and you have to zoom in to see the differences. As AMD points out, zoomed in images are not what a gamer will typically see during gameplay.

The Balanced preset was also very close to the target of 1440p but there were some differences noticeable compared to a normally rendered 1440p frame, especially around the image edges. During gameplay, it will be hard to tell the difference as you zip around, and you will need to stand still to perceive it.

The Performance preset once again performed great, getting really close to the target 1080p resolution but at a noticeable loss in image quality around the borders. It also seemed a smidge less sharp to our eyes.

Conclusion

Overall, FSR 2.0 delivers on its promises of better performance at barely any visual cost at the Quality preset. Even the Performance mode delivers great results, and it will be interesting to see how it fares against Nvidia's upcoming DLSS updates.